A Hierarchical Assessment of Adversarial Severity

TL;DR

This paper introduces a hierarchical approach to evaluate and improve the severity of adversarial attacks on neural networks, using a large-scale dataset and novel training methods to enhance robustness and reduce semantic errors.

Contribution

It proposes hierarchical attacks and curriculum training methods that leverage label structure to improve adversarial robustness and severity assessment.

Findings

Hierarchical attacks effectively measure adversarial severity.

Hierarchical curriculum training improves model robustness.

Models trained with hierarchy reduce severity and increase robustness.

Abstract

Adversarial Robustness is a growing field that evidences the brittleness of neural networks. Although the literature on adversarial robustness is vast, a dimension is missing in these studies: assessing how severe the mistakes are. We call this notion "Adversarial Severity" since it quantifies the downstream impact of adversarial corruptions by computing the semantic error between the misclassification and the proper label. We propose to study the effects of adversarial noise by measuring the Robustness and Severity into a large-scale dataset: iNaturalist-H. Our contributions are: (i) we introduce novel Hierarchical Attacks that harness the rich structured space of labels to create adversarial examples. (ii) These attacks allow us to benchmark the Adversarial Robustness and Severity of classification models. (iii) We enhance the traditional adversarial training with a simple yet effective Hierarchical Curriculum Training to learn these nodes gradually within the hierarchical tree. We perform extensive experiments showing that hierarchical defenses allow deep models to boost the adversarial Robustness by 1.85% and reduce the severity of all attacks by 0.17, on average.